plugins/_browser/skills/browser-forms/SKILL.md
Use this skill for complex Browser form workflows where the page state may depend on selects, checkboxes, radios, file uploads, contenteditable fields, validation, or visual confirmation.
Start with browser:content to capture current refs, then use browser:detail on ambiguous fields before acting. Prefer ref-based form actions before coordinates.
Use select_option, set_checked, upload_file, type, type_submit, and submit for form interaction. Use coordinates only when no stable ref exists or the UI is intentionally canvas-like.
Use browser:screenshot plus vision_load when layout, visual validation, captcha-like UI, canvas content, or hidden state matters. Browser screenshots are not automatically loaded into model-visible history.
Verify after submission with browser:content, browser:state, or another explicit browser:screenshot plus vision_load.
Do not guess file paths for upload. Verify that every path exists before calling upload_file.